wanshuiyin/Auto-claude-code-research-in-sleep ? reverse-engineered prompt

I want to build an AI research assistant to help automate my machine learning workflow. It should act like a research partner. I'd give it a research paper and its code from GitHub, and it should analyze the paper to find its weaknesses. From there, it should brainstorm ideas to improve the original work, run new experiments to validate those ideas, and then help rewrite the paper with the new results.

Another huge part of the job is handling peer review. When I get reviewer comments back, I need it to read them all, understand the critiques, and draft a structured, convincing rebuttal for the conference.

Finally, once a paper is accepted, it would be great if it could also generate a slide deck for a presentation and a poster from the paper's content. The whole system should be really lightweight, maybe using just Markdown files as "skills" so I'm not locked into a heavy framework. I've heard using one AI to do the work and a separate one to critique it is a good approach to avoid blind spots, so let's do that.